EP-CFG: Energy-Preserving Classifier-Free Guidance

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Kai, Luan, Fujun, Bi, Sai, Zhang, Jianming
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916521667723264
author Zhang, Kai
Luan, Fujun
Bi, Sai
Zhang, Jianming
author_facet Zhang, Kai
Luan, Fujun
Bi, Sai
Zhang, Jianming
contents Classifier-free guidance (CFG) is widely used in diffusion models but often introduces over-contrast and over-saturation artifacts at higher guidance strengths. We present EP-CFG (Energy-Preserving Classifier-Free Guidance), which addresses these issues by preserving the energy distribution of the conditional prediction during the guidance process. Our method simply rescales the energy of the guided output to match that of the conditional prediction at each denoising step, with an optional robust variant for improved artifact suppression. Through experiments, we show that EP-CFG maintains natural image quality and preserves details across guidance strengths while retaining CFG's semantic alignment benefits, all with minimal computational overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2412_09966
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle EP-CFG: Energy-Preserving Classifier-Free Guidance
Zhang, Kai
Luan, Fujun
Bi, Sai
Zhang, Jianming
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Classifier-free guidance (CFG) is widely used in diffusion models but often introduces over-contrast and over-saturation artifacts at higher guidance strengths. We present EP-CFG (Energy-Preserving Classifier-Free Guidance), which addresses these issues by preserving the energy distribution of the conditional prediction during the guidance process. Our method simply rescales the energy of the guided output to match that of the conditional prediction at each denoising step, with an optional robust variant for improved artifact suppression. Through experiments, we show that EP-CFG maintains natural image quality and preserves details across guidance strengths while retaining CFG's semantic alignment benefits, all with minimal computational overhead.
title EP-CFG: Energy-Preserving Classifier-Free Guidance
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2412.09966